Mercurial > repos > PredictPDPS / file comparison
comparison: src/AlgorithmGreedy.jl
src/AlgorithmGreedy.jl
- changeset 36
- e4a8f662a1ac
- parent 35
- 74b1a9f0c35e
- child 37
- bba159cf1636
equal
deleted
inserted
replaced
| 35:74b1a9f0c35e | 36:e4a8f662a1ac |
|---|---|
| 1 #################################################################### | |
| 2 # Predictive online PDPS for optical flow with known velocity field | |
| 3 #################################################################### | |
| 4 | |
| 5 __precompile__() | |
| 6 | |
| 7 module AlgorithmGreedy | |
| 8 | |
| 9 identifier = "pdps_known_greedy" | |
| 10 | |
| 11 using Printf | |
| 12 | |
| 13 using AlgTools.Util | |
| 14 import AlgTools.Iterate | |
| 15 using ImageTools.Gradient | |
| 16 | |
| 17 using ..OpticalFlow: ImageSize, | |
| 18 Image, | |
| 19 pdflow!, | |
| 20 Greedy | |
| 21 | |
| 22 ######################### | |
| 23 # Iterate initialisation | |
| 24 ######################### | |
| 25 | |
| 26 function init_rest(x::Image) | |
| 27 imdim=size(x) | |
| 28 | |
| 29 y = zeros(2, imdim...) | |
| 30 Δx = copy(x) | |
| 31 Δy = copy(y) | |
| 32 x̄ = copy(x) | |
| 33 | |
| 34 return x, y, Δx, Δy, x̄ | |
| 35 end | |
| 36 | |
| 37 function init_iterates(xinit::Image) | |
| 38 return init_rest(copy(xinit)) | |
| 39 end | |
| 40 | |
| 41 function init_iterates(dim::ImageSize) | |
| 42 return init_rest(zeros(dim...)) | |
| 43 end | |
| 44 | |
| 45 ############ | |
| 46 # Algorithm | |
| 47 ############ | |
| 48 | |
| 49 function solve( :: Type{DisplacementT}; | |
| 50 dim :: ImageSize, | |
| 51 iterate = AlgTools.simple_iterate, | |
| 52 params::NamedTuple) where DisplacementT | |
| 53 | |
| 54 ################################ | |
| 55 # Extract and set up parameters | |
| 56 ################################ | |
| 57 | |
| 58 α, ρ = params.α, params.ρ | |
| 59 R_K² = ∇₂_norm₂₂_est² | |
| 60 γ = 1.0 | |
| 61 Λ = params.Λ | |
| 62 τ₀, σ₀ = params.τ₀, params.σ₀ | |
| 63 | |
| 64 τ = τ₀/γ | |
| 65 @assert(1+γ*τ ≥ Λ) | |
| 66 σ = σ₀*1/(τ*R_K²) | |
| 67 | |
| 68 println("Step length parameters: τ=$(τ), σ=$(σ)") | |
| 69 | |
| 70 ###################### | |
| 71 # Initialise iterates | |
| 72 ###################### | |
| 73 | |
| 74 x, y, Δx, Δy, x̄ = init_iterates(dim) | |
| 75 init_data = (params.init == :data) | |
| 76 | |
| 77 #################### | |
| 78 # Run the algorithm | |
| 79 #################### | |
| 80 | |
| 81 v = iterate(params) do verbose :: Function, | |
| 82 b :: Image, | |
| 83 v_known :: DisplacementT, | |
| 84 🚫unused_b_next :: Image | |
| 85 | |
| 86 ################## | |
| 87 # Prediction step | |
| 88 ################## | |
| 89 if init_data | |
| 90 x .= b | |
| 91 init_data = false | |
| 92 end | |
| 93 | |
| 94 pdflow!(x, Δx, y, Δy, v_known, Greedy()) | |
| 95 | |
| 96 ############ | |
| 97 # PDPS step | |
| 98 ############ | |
| 99 | |
| 100 ∇₂ᵀ!(Δx, y) # primal step: | |
| 101 @. x̄ = x # | save old x for over-relax | |
| 102 @. x = (x-τ*(Δx-b))/(1+τ) # | prox | |
| 103 @. x̄ = 2x - x̄ # over-relax | |
| 104 ∇₂!(Δy, x̄) # dual step: y | |
| 105 @. y = (y + σ*Δy)/(1 + σ*ρ/α) # | | |
| 106 proj_norm₂₁ball!(y, α) # | prox | |
| 107 | |
| 108 ################################ | |
| 109 # Give function value if needed | |
| 110 ################################ | |
| 111 v = verbose() do | |
| 112 ∇₂!(Δy, x) | |
| 113 value = norm₂²(b-x)/2 + params.α*γnorm₂₁(Δy, params.ρ) | |
| 114 value, x, [NaN, NaN], nothing | |
| 115 end | |
| 116 | |
| 117 v | |
| 118 end | |
| 119 | |
| 120 return x, y, v | |
| 121 end | |
| 122 | |
| 123 end # Module | |
| 124 | |
| 125 |
